基于单目视觉的头部姿态估计算法的研究与实现

发布时间：2019-03-14 18:37

【摘要】：头部姿态估计是从图像或视频中估计出头部相对于摄像机的三个方位参数,即俯仰角(pitch)、偏航角(yaw)、翻滚角(roll)。该技术在监控系统,汽车安全辅助驾驶系统,游戏娱乐等领域有广泛应用,基于此,本文对单目视觉下的头部姿态估计问题进行了深入的研究。本文的主要工作如下:1.详细论述了整个头部姿态估计流程所涉及的关键算法,包括人脸检测、面部特征点模型以及姿态参数估计。2.原始的POSIT算法是从几何角度进行地推导,本文从代数角度,对POSIT算法进行了新的更加简洁地推导。3.重点研究了两种通用单目姿态估计算法---POSIT算法和EPNP算法在头部姿态估计中的应用,给出了比较详细的实验对比分析,分别研究了算法对于噪声、点数、不同角度下的参数变化的影响,结果可以看出POSIT算法在精度上略逊于EPNP算法,但前者在速度上具有较显著的优势,因此在实际问题中,需要在速度和精度上进行权衡来选择合适的算法。此外,本文搭建了一个完整的头部姿态估计在线演示系统。
[Abstract]:The head attitude estimation is to estimate the three azimuth parameters of the head relative to the camera from the image or video, namely the pitch angle (pitch), yaw angle (yaw), roll angle (roll). This technology has been widely used in the field of monitoring system, automobile safety assistant driving system, game entertainment and so on. Based on this, the problem of head attitude estimation under monocular vision is deeply studied in this paper. The main work of this paper is as follows: 1. The key algorithms involved in the whole head pose estimation process are discussed in detail, including face detection, facial feature point model and attitude parameter estimation. The original POSIT algorithm is derived from the geometric point of view. In this paper, the POSIT algorithm is derived more succinctly from the algebraic point of view. This paper focuses on the application of two general monocular attitude estimation algorithms-POSIT algorithm and EPNP algorithm in head attitude estimation. The results show that the accuracy of the POSIT algorithm is slightly lower than that of the EPNP algorithm, but the former has significant advantages in speed, so in practical problems, A trade-off between speed and accuracy is needed to select the appropriate algorithm. In addition, a complete on-line demonstration system for head attitude estimation is built in this paper.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】